Are Pretrained Language Models Symbolic Reasoners over Knowledge?

Proceedings of the 24th Conference on Computational Natural Language Learning, pages 552–564 Online, November 19-20, 2020. c⃝2020 Association for Computational Linguistics https://doi.org/10.18653/v1/P17 552 Are Pretrained Language Models Symbolic Reasoners Over Knowledge? Nora Kassner∗, Benno Krojer∗, Hinrich Sch¨utze Center for Information and Language Processing (CIS) LMU Munich, Germany kassner@cis.lmu.de Abstract How can pretrained language models (PLMs) learn factual knowledge from the training set? We investigate the two most important mech- anisms: reasoning and memorization. Prior work has attempted to quantify the number of facts PLMs learn, but we present, using syn- thetic data, the first study that investigates the causal relation between facts present in train- ing and facts learned by the PLM. For reason- ing, we show that PLMs seem to learn to apply some symbolic reasoning rules correctly but struggle with others, including two-hop rea- soning. Further analysis suggests that even the application of learned reasoning rules is flawed. For memorization, we identify schema conformity (facts systematically supported by other facts) and frequency as key factors for its success. 1 Introduction Pretrained language models (PLMs) like BERT (Devlin et al., 2019), GPT-2 (Radford et al., 2019) and RoBERTa (Liu et al., 2019) have emerged as universal tools that capture a diverse range of lin- guistic and – as more and more evidence suggests – factual knowledge (Petroni et al., 2019; Radford et al., 2019). Recent work on knowledge captured by PLMs is focused on probing, a methodology that identi- fies the set of facts a PLM has command of. But little is understood about how this knowledge is acquired during pretraining and why. We analyze the ability of PLMs to acquire factual knowledge focusing on two mechanisms: reasoning and mem- orization. We pose the following two questions: a) Symbolic reasoning: Are PLMs able to infer knowledge not seen explicitly during pretraining? b) Memorization: Which factors result in success- ful memorization of a fact by PLMs? ∗*equal contribution We conduct our study by pretraining BERT from scratch on synthetic corpora. The corpora are com- posed of short knowledge-graph like facts: subject- relation-object triples. To test whether BERT has learned a fact, we mask the object, thereby gener- ating a cloze-style query, and then evaluate predic- tions. Symbolic reasoning. We create synthetic cor- pora to investigate six symbolic rules (equivalence, symmetry, inversion, composition, implication, negation); see Table 1. For each rule, we create a corpus that contains facts from which the rule can be learned. We test BERT’s ability to use the rule to infer unseen facts by holding out some facts in a test set. For example, for composition, BERT should infer, after having seen that leopards are faster than sheep and sheep are faster than snails, that leopards are faster than snails. Our setup is similar to link prediction in the knowledge base domain and therefore can be seen as a natural extension of the question: “Language models as knowledge bases?” (Petroni et al., 2019). In the knowledge base domain, prior work (Sun et al., 2019; Zhang et al., …

Content truncated to 500 words